Graph2Structure

Graph2Structure (G2S) is a kernel ridge regression based machine learning model to predict 3D atomic structures from graph based information (e.g. SMILES).

Installation

Install all requirements using pip

pip install -r requirements.txt

To use G2S as a regular Python library, you have to add the repository to your $PYTHONPATH

export PYTHONPATH="/home/dominik/github/graph2structure"

Additional packages required to use features such as graph-extraction include

rdkit
xyz2mol
xTB

Distance Geometry

To reconstruct 3D coordinates from a distance matrix, you have to install a distance geometry solver such as DGSOL.

DGSOL is available at http://www.mcs.anl.gov/~more/dgsol/dgsol-1.3.tar.gz

Compile the software according to the README https://www.mcs.anl.gov/~more/dgsol/README and add the path to the dgsol binary to your `$PATH

Usage

For usage examples, take a look at the tutorials in the example folder!

References

@article{Lemmg2s2021,
  optdoi = {10.1038/s41467-021-24525-7},
  opturl = {https://optdoi.org/10.1038/s41467-021-24525-7},
  year = {2021},
  month = jul,
  publisher = {Springer Science and Business Media {LLC}},
  volume = {12},
  number = {1},
  author = {Dominik Lemm and Guido Falk von Rudorff and O. Anatole von Lilienfeld},
  title = {Machine learning based energy-free structure predictions of molecules,  transition states,  and solids},
  journal = {Nature Communications}
}